the importance of housing, accessibility, and ... paper.pdf · 39 constructs will significantly...

THE IMPORTANCE OF HOUSING, ACCESSIBILITY, AND TRANSPORTATION 1

CHARACTERISTIC RATINGS ON STATED NEIGHBORHOOD PREFERENCE 2

3 4

5

Steven R. Gehrke * 6

Department of Civil and Environmental Engineering 7

Portland State University 8

PO Box 751 9

Portland, OR 97207.0751 10

Email: [email protected] 11

12

Kristina M. Currans 13



16

Kelly J. Clifton, Ph.D. 17



20

21

June 1, 2015 22

23

24

Presentation at the 25

95th Annual Meeting of the Transportation Research Board 26

Washington, D.C. 27

January 10-14, 2016 28

29

30

Word Count: 160 Words (Abstract) + 4,544 Words (Paper) 31

Tables and Figures: 2,250 Words (7 Tables & 2 Figures) 32

Total Word Count: 6,954 Words (Under 35 References) 33

34

35

* Corresponding Author 36

37

mailto:[email protected]

Gehrke, Currans, & Clifton 2

ABSTRACT 1 The residential location decision process is predicated on transportation, accessibility, and housing 2

characteristics offered by a neighborhood. Consequently, understanding how these myriad 3

characteristics are valued by individuals will inform urban planners and decision makers of their 4

connection to neighborhood preference. Unfortunately, forecasting methods commonly lack the 5

specificity to recognize how dynamic residential environment preferences impact future land use 6

and transportation decisions. Too often, these policy instruments rely on socioeconomic 7

characteristics to measure heterogeneity in revealed location decisions. Using stated preference 8

data collected in Portland, Oregon, this study employed structural equation modeling techniques 9

to examine the influence of these socioeconomic measures and latent constructs of single-family 10

dwelling and non-automotive access importance on stated neighborhood preference. Model results 11

suggested that conventional socioeconomic measures insufficiently explain preference variation 12

and that the two latent constructs were robust predictors of an urban neighborhood preference. 13

This study’s findings emphasize the importance of utilizing characteristic ratings and conventional 14

socioeconomic measures in future stated neighborhood preference analyses. 15


BACKGROUND 1 The residential location decision-making process has long been an area of investigation in travel 2

behavior research. Such studies generally evaluate the connection between revealed land use or 3

location choices and travel outcomes. The goal has tended to focus on a better representation of 4

household taste variations in travel demand models, understanding of household self-selection into 5

residential environments supportive of their preferred travel behaviors, and explanation of trade-6

offs in housing location and job accessibility faced by households. Yet, despite improved 7

knowledge gained in these topical areas, travel demand applications and behavioral models still 8

commonly lack the specificity needed to explain how shifting residential environment preferences 9

will influence future housing, land use, and transportation decisions. Often, these policy 10

instruments solely rely on sociodemographic and economic characteristics (e.g., household size) 11

to measure differences in observed residential location choices; when in fact, neighborhood 12

preference explained by certain housing or travel attitudes (e.g., pro-transit) may precede the 13

choice itself (1). Findings from past studies using only socioeconomic status to observe revealed 14

residential location choice variation may therefore prove insufficient in advising the future demand 15

for urban neighborhoods because economic conditions and the housing and travel preferences of 16

emerging market segments are dynamic. 17

In response, an increased importance to create toolkits capable of providing practitioners 18

with robust and flexible behavioral models for understanding how stated preferences for future 19

residential environments vary within and between individuals has been placed on travel behavior 20

researchers. To address this identified need, our study examines stated neighborhood preference 21

variation predicted by both socioeconomic and subjective measurements to better understand the 22

influence of these determinants in future models of the residential location decision process. 23

Specifically, a stated preference survey was administered to (i) define those housing, accessibility, 24

and transportation characteristics most important to an individual’s residential location decision 25

and (ii) test whether importance ratings of these characteristics are stronger predictors of stated 26

neighborhood preference than conventional socioeconomic measures. The first hypothesis being 27

that the rated importance of these characteristics bundle together as latent constructs reflecting an 28

individual’s desire for certain housing, accessibility, and transportation features and amenities 29

offered by a neighborhood. The second hypothesis is that the importance of these characteristics 30

is a stronger determinant of unconstrained stated neighborhood preference than an individual’s 31

present socioeconomic circumstance. By distinguishing how the rated importance of residential 32

location characteristics bundle together as latent constructs and comparing their relative impact on 33

stated neighborhood preference to that of socioeconomic condition, this study provides fresh 34

insight into the complex residential location decision-making process. 35

Past stated preference studies have theorized the residential location decision process to be 36

most influenced by housing, accessibility, and transportation characteristics (2). In support, a 37

recent literature review summarized past studies as typically modeling a decision maker’s 38

preference in dwelling unit, location, and accessibility as well as socioeconomic status 39

characteristics (3). Other studies have suggested the added benefit of attitudinal measurement in 40

explaining residential location preferences for urban and suburban environments (4, 5). 41

Arguably, the most explored characteristic in stated residential location choice studies has 42

been dwelling type; however, past studies have also examined physical housing attributes related 43

to property size, dwelling size, and privacy. Hunt (6) and Molin and Timmermans (7) noted most 44

survey participants preferred single-family detached units when presented an array of distinct 45

dwelling type alternatives. Corroborating these findings, Senior and colleagues (8) found that 46


homeowners favored detached and semi-detached dwelling types over terraced homes or 1

apartments. Beyond a singular importance in dwelling type, Olaru et al (9) found a latent construct 2

also reflecting property size significantly predicted residential environment choice. Studying 3

image inclusion in preference surveys, Jansen and colleagues (10) found that having a detached 4

house, large dwelling and property size, and limited contact with neighbors significantly 5

influenced residential environment preference. 6

Characteristics of the physical environment surrounding a dwelling unit and accessibility 7

afforded by its proximity to local activities (e.g., shopping) have also been frequently studied. One 8

strategy to explain the physical environment variation has been neighborhood classification along 9

a continuum illustrating the relative distance to, density of, and diversity in accessible activity 10

locations (8). Neighborhood dissonance studies often dichotomize the residential environment as 11

either urban or rural (11, 12); whereas, other studies have assessed the presence of a single activity 12

location such as a local park (2, 10), quality in a nearby service such as a school (13), or distance 13

to urban facility such as a medical center (9, 14). Transportation facility access has also been 14

regularly used to measure variation in stated neighborhood preference. Regarding vehicle 15

infrastructure, Hunt (6) found that individuals preferred local streets in front of their dwelling unit 16

over collector roads and local streets with speed bumps; while, Senior and colleagues (8) and 17

Walker & Li (2) found no meaningful difference in residential preferences for on- and off-street 18

vehicle parking facilities. As for non-automotive transport, transit access was noted as an 19

influential determinant of residential location choice by Lund (15) and Hoshino (14). Similarly, 20

Olaru and colleagues (9) found relocation decisions were influenced by stated interest in transit 21

station proximity; likewise, most individuals considered walking and bicycling access as an 22

important characteristic. By contrast, Walker & Li (2) found bike path access had no significant 23

influence on residential location. 24

Finally, a recent stream of travel behavior research has evaluated revealed neighborhood 25

decisions, supplemented with attitudinal measures, to better understand the residential location 26

decision-making process. Khattak and Rodriguez (16) found the stated importance of nearby shops 27

and services, adjacency to sidewalks, and proximity to neighbors positively influenced the current 28

decision to reside in a neo-traditional neighborhood. Using structural equation modeling, Bagley 29

and Mokhtarian (4) found increased household size, respondent age, and scores for a set of latent 30

attitudinal constructs (e.g., pro-alternatives) positively predicted the revealed choice of a 31

traditional neighborhood. However, revealed neighborhood had a negligible effect on travel 32

behavior after accounting for socioeconomic and attitudinal measures. Such attitudinal measures 33

help to explain heterogeneity in location choices not exclusively accounted for by socioeconomic 34

status (2, 9, 12). This study extends the evidence base by explaining variation in the stated 35

preference for residential environments with both socioeconomic and attitudinal measurement. 36

Drawing from past research, we theorize that latent constructs will be reflected by items intended 37

to assess the importance of housing, accessibility, and transportation characteristics, and that these 38

constructs will significantly contribute to stated neighborhood preference. 39

METHODS 40

Research Design and Data Collection 41 An online cross-sectional survey of a probability sample was used to determine the importance of 42

housing, accessibility, and transportation characteristic ratings on stated neighborhood 43

preferences. The survey instrument had three sections designed to (i) collect participant 44

background information, (ii) measure the importance level a participant places on a set of 45


characteristics possibly impacting his/her decision process, and (iii) conduct a choice-based 1

conjoint experiment of neighborhood- and commute-based trade-offs (not analyzed in this study). 2

In the first section, participants provided information about their socioeconomic status and current 3

transportation and housing circumstances. To understand participants’ preferences for various 4

types of residential environments, a set of visual images with descriptive text was utilized to 5

provide a richer illustration of the physical characteristics for the four neighborhood concepts 6

(FIGURE 1). Participants were then asked to examine 17 items related to different housing, 7

accessibility, and transportation characteristics and rate the importance level (very important, 8

somewhat important, or not at all important) that each characteristic has in his/her residential 9

location decision-making process. 10

Participants were recruited from Portland, Oregon in two waves by mailing postcards with 11

a link to the survey website. In the first wave, postcards were mailed to 8,000 randomly selected 12

individuals from 201,444 home addresses in the metropolitan region during June 2014. In 13

November 2014, a second wave of postcards was sent to 1,982 downtown residents. Participants 14

of the Neighborhood Transportation Study were invited to enter their name for a chance to win a 15

gift card to an electronic commerce company. Response rate was 6.3% for the first recruitment 16

wave and 8.1% for the second wave. 17

TABLE 1 Descriptive statistics for sociodemographic and economic variables in sample and study area 18

Observed Variable

Sample Relative Portland Metro Region

n % Difference n %

Household Size

1 member 132 0.24 0.80 180,454 0.30

2 members 212 0.38 1.12 207,758 0.34

3 members 105 0.19 1.27 92,001 0.15

4 or members 105 0.19 0.90 126,677 0.21

Annual Household Income

$0 - $24,999 62 0.12 0.57 124,519 0.21

$25,000 - $49,999 113 0.21 0.88 143,007 0.24

$50,000 - $99,999 185 0.35 1.09 194,782 0.32

$100,000 or more 174 0.33 1.38 144,582 0.24

Respondent Age

18 - 34 years 145 0.26 0.79 382,343 0.33

35 - 44 years 117 0.21 1.05 229,130 0.20

45 - 64 years 209 0.38 1.12 395,560 0.34

65 or more years 78 0.14 1.00 163,587 0.14

Respondent Gender

Female 276 0.50 0.98 600,138 0.51

Male 267 0.48 0.98 570,482 0.49

Note: Portland Metro Region summary statistics were calculated from 2007-2011 American Community Survey.

Survey Participants 19

In total, 654 participants from the Portland metropolitan region completed the survey. The study 20

sample consisted of 554 individuals, who provided sufficient information to examine the measures 21

of interest. TABLE 1 summarizes select socioeconomic measures for the study sample. Inspection 22

of stratified levels revealed a majority of participants reported a household size of 2 members, with 23

a similar sample proportion having a household size of three or more members. One-third of the 24

sample earned an annual household income from $50,000-$99,999. A majority of participants 25

reported being over 45 years old, while half the sample stated they were female. Although average 26


household size of the sample (2.46) was equivalent to the study area, a further comparison of the 1

sample to the metropolitan region revealed the average survey participant was more likely to earn 2

a higher annual household income and be older than the average Portland resident. 3

4

FIGURE 1 Preferred neighborhood concepts in the Neighborhood Transportation Study 5

Measures 1 In the survey instrument, visual collages of nine images complemented with text descriptions 2

exemplified four independent concepts of neighborhood types: central district, urban residential 3

district, urban neighborhood, and suburban neighborhood. Images selected to visualize the 4

concepts were collected from an assortment of Google Street View (17) screenshots. Image 5

location was chosen from an aggregation of objectively defined neighborhood types based on 6

measurement of the activity density, employment entropy, and intersection density of the built 7

environment in the study area. Detailed description of the overall visualization process, comprising 8

the objective definition of neighborhood types, adoption criteria for image selection, and validation 9

of neighborhood concepts is offered elsewhere (18). 10

Neighborhood preference was measured by the participant’s response to a question asking 11

him/her to examine four image collages with text descriptions and select the neighborhood where 12

he/she would most prefer to live. Text descriptions illustrated a bundle of neighborhood attributes 13

related to the typical dwelling type and living space; likely home ownership status; distance to 14

retail, services, and entertainment activities; vehicle parking availability; and public transit access 15

to regional centers that were objectively measured and visualized to be at different levels across 16

the concepts. Each neighborhood exists along an urban continuum. In the study sample, 49 (9%) 17

respondents preferred the central district, 99 (18%) preferred an urban residential district, 220 18

(40%) preferred an urban neighborhood, and the remaining 186 (34%) respondents preferred the 19

suburban neighborhood concept. 20

TABLE 2 Importance ratings for various housing, accessibility, and transport characteristics 21

Observed Characteristic

Level of Importance

Very Somewhat Not at All

n n % n % n %

1: Owning a house/condo 540 343 0.64 133 0.25 64 0.12

2: Living in a home with a large living space 534 137 0.26 226 0.42 171 0.32

3: Living in a detached single-family home 536 233 0.43 172 0.32 131 0.24

4: Having a private yard 539 262 0.48 175 0.32 102 0.19

5: Having privacy from my neighbors 535 278 0.52 227 0.42 30 0.06

6: Living at the 'center of it all' 538 81 0.15 213 0.40 244 0.45

7: Having access to highways/freeways 534 119 0.22 271 0.51 144 0.27

8: Having a variety of transportation options 540 251 0.46 228 0.42 61 0.11

9: Walking to bus and/or rail stops 542 257 0.47 186 0.34 99 0.18

10: Having off-street parking at local destinations 529 98 0.19 263 0.50 168 0.32

11: Having dedicated parking at your residence 540 362 0.67 115 0.21 63 0.12

12: Having access to parks and recreational areas 541 320 0.59 196 0.36 25 0.05

13: Walking to nearby places 538 352 0.65 146 0.27 40 0.07

14: Biking to nearby places 536 155 0.29 185 0.35 196 0.37

15: Being near high-quality public schools 536 152 0.28 110 0.21 274 0.51

16: Living near established, older homes 531 76 0.14 188 0.35 267 0.50

17: Having a commute that takes 25 minutes or less 538 320 0.59 147 0.27 71 0.13

Note: The most frequently stated level of importance for each characteristic is shown in bold. 22

After selecting a preferred neighborhood, participants were then asked to examine a set of 23

17 housing, accessibility, and transport characteristics and state whether each item was very, 24

somewhat, or not at all important. TABLE 2 shows the importance level that respondents selected 25

for each residential location characteristic. Examination of survey responses revealed that sampled 26

residents placed the greatest importance on having dedicated parking at their residence (67%), 27

walking to nearby places (65%), owning their residence (64%), having a 25 minute or less 28


commute (59%), and accessing parks and recreational areas (59%). In contrast, respondents stated 1

being near high-quality public schools (51%), living near older homes (50%), living at the ‘center 2

of it all’ (45%), biking to nearby places (37%), and living in a home with a large living space 3

(32%) bared no importance on their residential decision making process. 4

Statistical Analyses 5 A two-part analytic plan composed of (i) exploratory factor analysis (EFA) of rated importance 6

characteristics and (ii) structural equation model (SEM) to determine associations between latent 7

importance rating constructs and stated neighborhood concept preference was employed. The EFA 8

technique was used to help generate a theoretic understanding of the internal structure of how 9

observed importance characteristic ratings may improve construct measurement (19). An 10

assumption being that factors shaped by this exploratory technique may also be useful as 11

operational descriptions. To test whether these operational descriptions also represent theoretical 12

constructs, SEM was then used—informed by both the EFA results and past literature of how 13

various characteristics may bundle together—to predict stated neighborhood preference. Sample 14

size restrictions prohibited splitting the collected data into subsamples to independently conduct 15

the two analyses. 16

EFA was performed in sequential steps centered on three decisions related to selection of a factor 17

model approach, extraction scheme, and rotation method (20). Principal components analysis was 18

selected as a modeling strategy to form uncorrelated linear combinations of the importance rating 19

characteristics. Inspection of eigenvalues associated with each resulting factor and their scree plot 20

display guided the factor extraction (21). Principal axis factoring was employed since this method 21

has generally outperformed other extraction methods in recovering factors with low loadings, 22

providing solutions with stable loadings, and isolating correlated factors (22). Promax rotation, 23

which allows correlation between extracted factors, was adopted as a rotation method leading to a 24

final factor model. 25

Items loading in the chosen factor solution were assessed in tandem with theory to identify 26

latent construct indicators related to the stated importance of housing, accessibility, and 27

transportation characteristics in residential location decisions. Using a two-step approach, a 28

measurement model positing a relationship between a set of these observed characteristics to a 29

latent construct(s) was estimated through confirmatory factor analysis (CFA) prior to assessing a 30

structural model with path assignments (23). Structural equation modeling has been firmly 31

established as an analytic strategy in which a set of equations consisting of a measurement model 32

for exogenous variables may be concurrently estimated in a structural model identifying 33

associations between endogenous and exogenous variables. An SEM analysis has several 34

advantages over conventional multivariate regression methods, including an ability to: (i) 35

simultaneously test direct/indirect effects and bidirectional relationships across different paths, (ii) 36

correct for measurement error in observed variables, and (iii) assess latent constructs of multiple 37

indicators (24). Accordingly, adopting SEM enabled latent constructs reflecting the importance of 38

residential location characteristic ratings to be simultaneously estimated free of measurement error 39

and predictive of stated neighborhood preference, while accounting for direct and indirect paths 40

with self-reported socioeconomic variables. 41


RESULTS 1

Exploratory Factor Analysis 2 Correlations between importance characteristics were analyzed prior to the EFA. The importance 3

characteristics, reflecting the participant rating of 17 items describing their residential location 4

decision process, were coded 3 for very important, 2 for somewhat important, and 1 for not at all 5

important. TABLE 3 shows a zero-order correlation matrix of the 14 retained rating items. 6

Measures linked to being near high-quality public schools, living near older homes, and having a 7

25 minute or less commute were removed from this, and subsequent, analyses due to either item 8

misconstruction or lack of item response variation within the study sample. 9

TABLE 3 Zero-order correlation matrix of importance rating characteristics 10

Observed Importance Rating Measure 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1: Owning a house/condo ---

2: … home with a large living space .18 ---

3: Living in a detached single-family home .34 .26 ---

4: Having a private yard .28 .26 .68 ---

5: Having privacy from my neighbors .17 .20 .35 .33 ---

6: Living at the 'center of it all' -.16 -.14 -.43 -.40 -.34 ---

7: Having access to highways/freeways .19 .18 .11 .07 .12 -.11 ---

8: … variety of transportation options -.22 -.20 -.36 -.31 -.27 .33 -.11 ---

9: Walking to bus and/or rail stops -.28 -.17 -.41 -.36 -.25 .32 -.14 .57 ---

10: … off-street parking … .12 .17 .14 .11 .19 -.11 .25 -.15 -.18 ---

11: … dedicated parking at your residence .32 .18 .27 .23 .26 -.23 .30 -.27 -.25 .36 ---

12: … access to parks … -.07 -.08 -.11 -.09 -.10 .10 -.02 .23 .13 -.01 -.05 ---

13: Walking to nearby places -.23 -.22 .40 -.33 -.30 .41 -.16 .50 .50 -.16 -.30 .24 ---

14: Biking to nearby places -.08 -.15 -.04 -.04 -.13 .06 -.16 .24 .18 -.12 -.19 .32 .31 ---

Note: Kendall rank correlation coefficients over 0.40 or under -0.40 appear in bold. 11

An EFA was next conducted to advise the development of theoretical constructs related to 12

the importance of housing, accessibility, and transportation characteristics to residential location 13

decisions. Provided the subjectivity of EFA, TABLE 4 shows three possible factor solutions in 14

line with the rule of eigenvalues above one (25). The two-factor solution represented the best 15

balance between a priori theory and an empirical description of the data (χ2 (64) = 292.75, p < 16

0.01). Four items on Factor 2A with salient loadings described the transportation characteristics 17

related to the residential location decision-making process; whereas, each of the three high loading 18

items on Factor 2B conceptually reflected housing and accessibility features. Factor 2A was driven 19

by importance in having a variety of transportation options, walking to nearby places, walking to 20

bus and/or rail stops, biking to nearby places, and having access to parks and recreational areas. In 21

turn, the items reflecting Factor 2B included the importance of living in a detached single-family 22

home, having a private yard, and living at the ‘center of it all.’ In general, the eight items in the 23

two-factor solution with strong factor loaded on a distinct factor with the clear exception being the 24

importance of biking to nearby places. Also of note, living at the ‘center of it all’ was the only 25

characteristic with a strong negative loading. All seven items with strong loadings in the two-factor 26

solution were retained for the CFA as were those items with a factor loading greater than 0.30 or 27

less than -0.30 on either factor. 28

TABLE 4 Summary of three exploratory factor analyses for importance rating characteristics 1

Factor Solution: One Two Three

Factor Number: 1A 2A 2B 3A 3B 3C

Observed Importance Rating Measure

1: Owning a house/condo 0.42 -0.16 0.30 0.24 0.33

2: … home with a large living space 0.38 -0.20 0.22 0.19 0.21

3: Living in a detached single-family home 0.73 0.91 0.88

4: Having a private yard 0.66 0.87 0.85

5: Having privacy from my neighbors 0.49 -0.21 0.32 -0.12 0.29 0.17

6: Living at the 'center of it all' -0.58 0.19 -0.44 0.21 -0.44

7: Having access to highways/freeways 0.25 -0.28 -0.11 0.57

8: … variety of transportation options -0.66 0.75 0.79

9: Walking to bus and/or rail stops -0.69 0.64 -0.13 0.65 -0.15

10: … off-street parking … 0.28 -0.29 0.57

11: … dedicated parking at your residence 0.46 -0.35 0.14 0.73

12: … access to parks … -0.24 0.42 0.13 0.46 0.11

13: Walking to nearby places 0.68 0.70 0.69

14: Biking to nearby places -0.27 0.63 0.33 0.50 0.31 -0.17

Eigenvalue 3.72 2.47 2.18 2.06 2.02 1.40

Percent of Variance Explained 0.27 0.18 0.16 0.15 0.14 0.10

Note: Factor loadings over 0.40 or under -0.40 appear in bold; Factor loadings between -0.10 and 0.10 not shown. 2

Structural Equation Model 3 Guided by the empirically-driven EFA results and a priori hypotheses concerning the relationship 4

among importance rating characteristics, a CFA model was specified for the two latent constructs 5

reflecting single-family dwelling and non- automotive access importance to the residential location 6

decision making process. This analysis was conducted in R (26) using the ‘lavaan’ package (27), 7

which enabled use of categorical variables with a robust weighted least squares mean- and 8

variance-adjusted (WLSMV) estimator. Choice of model fit indices and overall performance was 9

guided by recommendations of Hu and Bentler (28), while simulation results (29) established that 10

a sample size greater than 500 cases offered sufficient power to reject models with a WLSMV 11

estimator. 12

Two latent constructs measuring rated single-family dwelling and non-automotive access 13

importance were entered into the final CFA model (TABLE 5). Although, the model chi-square 14

was significant (χ2 (13) = 67.48, p < 0.01) and root mean square error of approximation (RMSEA) 15

was above 0.06; both the comparative fit index (CFI) and Tucker-Lewis Index (TLI) were above 16

0.95, supporting an acceptable model fit to the sample data. Items for each latent construct were 17

above an acceptable standardized loading (β ≥ 0.40). A significant, negative relationship existed 18

between these two latent constructs. 19

The final CFA model was identified as a baseline structure to test measurement invariance 20

of the latent constructs. Measurement invariance is centered on the notion of a measuring device 21

functioning differently across varied conditions or market segments, which subsequently produces 22

systematic measurement inaccuracy (30). The objective of a multigroup invariance was to 23

determine if measurement equivalence may be established prior to specification of the two 24

importance rating constructs in a path model predicting neighborhood preference. Multigroup 25

membership was identified by a median-split of the study sample based on household size, annual 26

income, and respondent age. Assessment of factorial invariance across the three divisions followed 27

the procedure for categorical measures introduced by Millsap and Yun-Tein (31). Configural 28

measurement invariance existed when the sample was divided by households with one or two 29


members versus those with at least three members. Similarly, configural invariance was found 1

when a respondent age of 45 years was used to split the sample. In terms of household income, 2

factor loadings were equivalent across groups (weak invariance) when the sample was divided by 3

those households earning more or less than $50,000. Although weak measurement invariance has 4

been reported as a tolerable condition in practice (32), these multigroup invariance test results 5

informed a decision to control for each of the observed HIA measures as socioeconomic covariates 6

in the final path model. 7

TABLE 5 Confirmatory factor analysis of stated importance constructs (n=548) 8

Parameter Estimates: B SE (B) β p-value

Construct Variables

Factor A: Single-family dwelling importance

3: Living in a detached single-family home 1.00 --- 0.95 ---

4: Having a private yard 0.54 0.17 0.86 0.00

6: Living at the 'center of it all' * 0.32 0.08 0.70 0.00

Factor B: Non-automotive access importance

8: Having a variety of transportation options 0.84 0.14 0.83 0.00

9: Walking to bus and/or rail stops 0.90 0.15 0.85 0.00

11: Having dedicated parking at your residence * 0.36 0.07 0.54 0.00

13: Walking to nearby places 1.00 --- 0.87 ---

Covariances

Factor A ~~ Factor B -3.96 1.04 -0.73 0.00

Note: Dashes (---) indicate the standard error was not estimated. A star (*) indicates the measure was reverse-coded. 9 χ2 (13) = 67.48, p = 0.00. CFI = 0.99, TLI = 0.98, and RMSEA = 0.08. 10 11

Prior to inclusion of the HIA measures, a path model predicting the impact of single-family 12

dwelling and non-automotive access importance on an individual’s stated preference for one of the 13

four neighborhood concepts was estimated (TABLE 6). Similar to CFA model results, alternative 14

fit indices suggested a satisfactory overall model fit to the data (χ2 (33) = 125.15, p < 0.01, CFI = 15

0.98, TLI = 0.97, RMSEA = 0.07). Single-family dwelling importance negatively predicted 16

preference in central district and urban residential district neighborhood concepts; however, non-17

automotive access importance had no significant influence on stated preference in these two most 18

urban neighborhood concepts. A direct effect analysis of the importance rating constructs on 19

suburban neighborhood preference revealed single-family dwelling importance had a positive 20

effect (β = 0.56, p < 0.01), while non-automotive access importance had a negative effect (β = -21

0.37, p < 0.01) on the binary outcome. Of particular interest, single-family dwelling (β = 0.66, p < 22

0.01) and non-automotive access (β = 0.75, p < 0.01) importance both had a strong positive effect 23

on urban neighborhood preference. This finding suggests a trade-off in single-family dwelling 24

importance may be occurring when an individual contemplates between an urban residential 25

district and urban neighborhood; while, non-automotive importance trade-offs may be occurring 26

when deciding between urban or suburban neighborhoods. 27

Complexity to the base SEM specification was tested by adding exogenous HIA variables 28

as both predictors of neighborhood preference and the importance rating constructs. Path model 29

specification was performed with a two-step backward elimination approach in which direct paths 30

into the latent constructs were iteratively removed until only significant paths (p < 0.05) remained. 31

Once all direct paths from the HIA variables to the latent constructs met the significance threshold, 32

then direct paths from the HIA measures and latent constructs to the binary stated neighborhood 33

preference outcome were iteratively removed beginning at the highest p-value until only 34

significant paths remained. TABLE 7 provides final model results for stated urban neighborhood 35


preference, which was reasoned to be the neighborhood at which participants were able to suffice 1

the competing importance constructs; while FIGURE 2 offers a path diagram for this model. Model 2

fit indices have suggested a good fit to the data (χ2 (76) = 109.37, p < 0.05, CFI = 0.96, TLI = 3

0.95, RMSEA = 0.03). 4

TABLE 6 Structural equation model of neighborhood preference with stated importance constructs (n=548) 5


Measurement Model

Construct Variables






8: Having a variety of transportation options 0.65 0.12 0.81 0.00



13: Walking to nearby places 1.00 --- 0.94 ---

Covariances

Factor A ~~ Factor B -4.11 0.94 -0.72 0.00

Structural Model

Central District ^ ~ Factor A -0.46 0.14 -0.79 0.00

Central District ^ ~ Factor B -0.01 0.12 -0.01 0.96

Urban Residential District ^ ~ Factor A -0.47 0.11 -0.84 0.00

Urban Residential District ^ ~ Factor B -0.10 0.09 -0.13 0.26

Urban Neighborhood ^ ~ Factor A 0.29 0.09 0.66 0.00

Urban Neighborhood ^ ~ Factor B 0.42 0.12 0.75 0.00

Suburban Neighborhood ^ ~ Factor A 0.41 0.10 0.56 0.00

Suburban Neighborhood ^ ~ Factor B -0.34 0.10 -0.37 0.00

Note: Dashes (---) indicate the standard error was not estimated. A star (*) indicates the measure was reverse-coded. 6 A carrot (^) indicates a binary measure. χ2 (33) = 125.15, p = 0.00. CFI = 0.98, TLI = 0.97, and RMSEA = 0.07. 7 8

The latent constructs of non-automotive access (β = 0.70, p < 0.05) and single-family 9

dwelling (β = 0.68, p < 0.01) importance strongly predicted urban neighborhood preference. The 10

direct effect of the theoretical constructs on neighborhood preference outweighed most traditional 11

socioeconomic indicators. Having a household with three members was found to have a significant 12

direct effect on urban neighborhood preference (β = -0.17, p < 0.05), as was having a household 13

with four or more members (β = -0.14, p < 0.05). All binary household size measures significantly 14

predicted the importance of a single-family dwelling to the residential location decision-making 15

process. Having an annual household income between $25,000 and $49,999 significantly predicted 16

a negative stated urban neighborhood preference (β = -0.14, p < 0.05). In regard to the direct effect 17

of income on the two latent constructs, membership in the lowest income bracket positively 18

predicted the importance of non-automotive access (β = 0.17, p < 0.05); whereas, annual household 19

income over $100,000 negatively predicted single-family dwelling importance (β = -0.12, p < 20

0.05). Affiliation with the youngest age cohort positively predicted stated urban neighborhood 21

preference (β = 0.14, p < 0.05), while this cohort membership negatively predicted single-family 22

dwelling (β = -0.22, p < 0.01) and positively predicted non-automotive access importance (β = 23

0.23, p < 0.05). Participants aged 65 or more years negatively predicted single-family dwelling 24

importance (β = -0.13, p < 0.05) in the residential location decision process. 25


TABLE 7 Structural equation model of urban neighborhood preference with stated importance constructs 1 and observed socioeconomic measures (n=530) 2


Measurement Model

Construct Variables






8: Having a variety of transportation options 1.00 --- 0.88 ---



13: Walking to nearby places 0.92 0.42 0.86 0.03

Covariances

Factor A ~~ Factor B -2.65 1.20 -0.66 0.03

Structural Model

Urban Neighborhood ^ ~ Factor A 0.28 0.11 0.68 0.01

Urban Neighborhood ^ ~ Factor B 0.52 0.22 0.70 0.02

Urban Neighborhood ^ ~ 3 members ^ -0.53 0.26 -0.17 0.04

Urban Neighborhood ^ ~ 4 or more members ^ -0.46 0.22 -0.14 0.04

Urban Neighborhood ^ ~ $25,000 - $49,999 ^ -0.42 0.19 -0.14 0.03

Urban Neighborhood ^ ~ 18 - 34 years ^ 0.40 0.18 0.14 0.03

Factor A ~ 1 member ^ -1.24 0.45 -0.18 0.01

Factor A ~ 3 members ^ 1.45 0.55 0.19 0.01

Factor A ~ 4 or more members ^ 2.60 0.80 0.34 0.00

Factor A ~ $100,000 or more ^ -0.77 0.35 -0.12 0.03

Factor A ~ 18 - 34 years ^ -1.52 0.50 -0.22 0.00

Factor A ~ 65 or more years ^ -1.16 0.47 -0.13 0.01

Factor B ~ 4 or more members ^ -1.07 0.45 -0.25 0.02

Factor B ~ $0 - $24,999 ^ 0.91 0.39 0.17 0.02

Factor B ~ 18 - 34 years ^ 0.87 0.35 0.23 0.01

Factor B ~ 35 - 44 years ^ 0.62 0.28 0.15 0.03

Note: Dashes (---) indicate the standard error was not estimated. A star (*) indicates the measure was reverse-coded. 3 A carrot (^) indicates a binary measure. χ2 (76) = 109.37, p = 0.01. CFI = 0.96, TLI = 0.95, and RMSEA = 0.03. 4


1

FIGURE 2 Path diagram of urban neighborhood preference with stated importance constructs and observed 2 socioeconomic measures (n=530) 3

DISCUSSION 4 The primary contributions of this study were twofold. One contribution was an assessment of the 5

relative importance for different housing, accessibility, and transport characteristic ratings as they 6

relate to stated neighborhood preference. Using exploratory and confirmatory factor analysis, two 7

latent constructs describing the stated importance of single-family dwelling and non-automotive 8

access were identified. A potential third construct related to stated automotive access importance 9

was explored, but not confirmed. While not mutually exclusive, these latent constructs appear to 10

represent competing standpoints of the residential environment characteristics that mattered most 11

to the sampled residents. Study findings also hinted at an ordering of these housing and transport 12

attribute bundles in the stated preference for certain neighborhoods along an urban continuum. 13

Examination of the traced unstandardized effects (33) of the constructs to each 14

neighborhood concept in the base SEM estimation revealed single-family dwelling importance 15

was positively related to suburban neighborhood preferences, but became increasingly negative as 16

respondents’ stated preference increased along the urban continuum. Conversely, stated non-17

automotive access importance negatively predicted suburban preferences and became ever more 18

positive with an increase along this neighborhood continuum. The competing latent constructs 19

both negatively predicted a stated preference for urban neighborhoods; indicating this concept may 20

have reflected the residential environment in which survey participants faced a concession in their 21

unconstrained preference. This trend remained, but with a positive standardized relationship, after 22

controlling for the observed socioeconomic circumstance in the path model. 23


Reflection on factor construction revealed opportunities for refinement through an addition 1

of housing, accessibility, and transport characteristics to the survey instrument. For instance, the 2

highest indicator of Factor A was the importance of living in a detached single-family home. Is 3

this result based solely on building structure importance or an artefact of living space requirement, 4

some long-term goal of financial stability in retirement, or any other intrinsic meaning tied to the 5

significance in having this specific dwelling type? Clearer insight may also be gained by dividing 6

the accessibility item of living at the ‘center of it all’ into items expressing proximity to specific 7

activity locations (e.g., restaurant, market). Finally, separating biking to nearby places into items 8

referring to utilitarian and recreational travel may have led the latter characteristic to be a strong 9

indicator of the non-automotive access importance construct. 10

A second study contribution was an evaluation of whether socioeconomic condition should 11

approximate an unexpressed importance within heterogeneous households for certain housing, 12

accessibility, and transport features in their preferred residential environment. Recent studies have 13

stressed the advantage of including subjective measures in neighborhood preference prediction 14

(12, 34). Travel behavior research has traditionally relied on socioeconomic measures to model 15

neighborhood preference variation; however, findings of this study contribute to the evidence base 16

by demonstrating that objective and subjective measures are required to understand this complex 17

decision making process (5). While household size, income, and respondent age contributed to 18

each importance rating construct within the final path model, these objective HIA measures were 19

generally less predictive than either latent construct in explaining urban neighborhood preference 20

variation. 21

When tracing the overall unstandardized impact of each socioeconomic characteristic to 22

urban neighborhood preference, the effect of Factor A remained larger than all socioeconomic 23

characteristics with the lone notable exception being that single-person households were less likely 24

to prefer an urban neighborhood. Inspection of the other socioeconomic characteristics revealed 25

nonlinear relationships with the two importance rating constructs. Tracing annual income effects 26

on non-automotive access importance indicated that individuals in households earning between 27

$50,000 and $99,999 tended to have the lowest scores on Factor B; whereas, individuals within 28

households in the lowest and highest annual income brackets expressed the greatest importance in 29

non-automotive access. Studying the traced unstandardized effect of age revealed that respondents 30

between 45 and 64 years tended to place the greatest importance on single-family dwelling living, 31

and that the youngest and oldest cohorts had a comparatively negative relationship with Factor A. 32

While new and informative, results of this study are limited by content and context. Tracing 33

results indicated nonlinearities in the impact of socioeconomic characteristics on neighborhood 34

preference; unfortunately, robust evaluation of interaction effects was limited by sample size. Also, 35

individual preferences were self-reported and likely inaccurately portrayed the joint residential 36

location preferences of a multimember household seeking to best satisfy the needs of all members. 37

Likewise, while structural models are valuable in identifying predictive paths between variables 38

of interest, a choice-based conjoint analysis would enable a rigorous test of residential environment 39

preference free of omitted variable bias (7). Future research efforts will evaluate constrained 40

neighborhood choice and neighborhood dissonance by estimating the impact of the latent 41

importance rating constructs in addition to the socioeconomic attributes within the context of an 42

individual’s present circumstance. An initial hypothesis is that conventional socioeconomic 43

characteristics play a more influential role in constrained choice of neighborhood, but that so-44

called soft measures illuminate some unexplained heterogeneity in emerging market segments. 45

This future research offers an opportunity to build on the contribution of this study that has 46


underlined the importance of attitudinal measurement in better understanding the complexities in 1

the residential location decision making process. 2

ACKNOWLEDGEMENTS 3 The authors are thankful for the financial support of the Oregon Department of Transportation, 4

National Institute for Transportation and Communities, and Dwight D. Eisenhower Transportation 5

Fellowship Program. 6

REFERENCES 7 1. Handy, S., P. Mokhtarian, T. Buehler, and X. Cao. Residential Location Choice and Travel 8

Behavior: Implications for Air Quality. Publication UCD-ITS-RR-04-48. University of 9

California Davis - Institute of Transportation Studies, 2004. 10

2. Walker, J., and J. Li. Latent Lifestyle Preferences and Household Location Decisions. 11

Journal of Geographic Systems, Vol. 9, 2007, pp. 77-101. 12

3. Schirmer, P. M., M. A. B. van Eggermon, and K. A. Axhausen. The Role of Location in 13

Residential Location Choice Models: A Review of Literature. Journal of Transport and 14

Land Use, Vol. 7, No. 2, 2014, pp. 3-21. 15

4. Bagley, M. N., and P. L. Mokhtarian. The Impact of Residential Neighborhood Type on 16

Travel Behavior: A Structural Equations Modeling Approach. The Annals of Regional 17

Science, Vol. 36, 2002, pp. 279-297. 18

5. Bohte, W., K. Maat, and B. van Wee. Measuring Attitudes in Research on Residential Self-19

Selection and Travel Behaviour: A Review of Theories and Empirical Research. Transport 20

Reviews, Vol. 29, No. 3, 2009, pp. 325-357. 21

6. Hunt, J. D. Stated Preference Analysis of Sensitivities to Elements of Transportation and 22

Urban Form. In Transportation Research Record: Journal of the Transportation Research 23

Board, No. 1780, Transportation Research Board of the National Academies, Washington, 24

D.C., 2001, pp. 13-20. 25

7. Molin, E., and H. Timmermans. Testing Hierarchical Information Integration Theory: The 26

Causal Structure of Household Residential Satisfaction. Environment and Planning A, Vol. 27

35, No. 1, 2003, pp. 43-58. 28

8. Senior , M. L., C. J. Webster, and N. E. Blank. Residential Relocation and Sustainable 29

Urban Form: Statistical Analyses of Owner-Occupiers' Preferences. International 30

Planning Studies, Vol. 11, No. 1, 2006, pp. 41-57. 31

9. Olaru, D., B. Smith, and J. Taplin. Residential Location and Transit-Oriented Development 32

in a New Rail Corridor. Transportation Research Part A, Vol. 45, 2011, pp. 219-237. 33

10. Jansen, S., H. Boumeester, H. Coolen, R. Goetgeluk, and E. Molin. The Impact of 34

Including Images in a Conjoint Measurement Task: Evidence from Two Small-Scale 35

Studies. Journal of Housing and the Built Environment, Vol. 24, No. 3, 2009, pp. 271-297. 36

11. Schwanen, T., and P. L. Mokhtarian. The Extent and Determinants of Dissonance between 37

Actual and Preferred Residential Neighborhood Type. Environment and Planning B, Vol. 38

31, 2004, pp. 759-784. 39


12. De Vos, J., B. Derudder, V. Van Acker, and F. Witlox. Reducing Car Use: Changing 1

Attitudes or Relocating? The Influence of Residential Dissonance on Travel Behavior. 2

Journal of Transport Geography, Vol. 22, 2012, pp. 1-9. 3

13. Morrow-Jones, H. A., E. G. Irwin, and B. Roe. Consumer Preference for Neotraditional 4

Neighborhood Characteristics. Housing Policy Debate, Vol. 15, No. 1, 2004, pp. 171-202. 5

14. Hoshino, T. Estimation Of The Preference Heterogeneity Within Stated Choice Data Using 6

Semiparametric Varying-Coefficient Methods. Empirical Economics, Vol. 45, No. 3, 2013, 7

pp. 1129-1148. 8

15. Lund, H.. Reasons for Living in a Transit-Oriented Development, and Associated Transit 9

Use. Journal of the American Planning Association, Vol. 72, No. 3, 2006, pp. 357-366. 10

16. Khattak, A. J., and D. Rodriguez. Travel Behavior in Neo-Traditional Neighobrhood 11

Developments: A Case Study in USA. Transportation Research Part A, Vol. 39, 2005, pp. 12

481-500. 13

17. Google, Inc. Google Maps Street View. 2011-2014. https://www.google.com/maps. 14

Accessed May 2014. 15

18. Currans, K. M., S. R. Gehrke, and K. J. Clifton. Visualizing Neighborhoods in 16

Transportation Surveys: Testing Respondent Perceptions of Housing, Accessibility, and 17

Transportation Characteristics. Presented at 94th Annual Meeting of the Transportation 18

Research Board, Washington, D.C., 2014. 19

19. Henson, R. K., and J. K. Roberts. Use of Exploratory Factor Analysis in Published 20

Research: Common Errors and Some Comment on Improved Practice. Educational and 21

Psychological Measurement, Vol. 66, No. 3, 2006, pp. 393-416. 22

20. Ford, J. K., R. C. MacCallum, and M. Tait. The Application of Exploratory Factor Analysis 23

in Applied Psychology: A Critical Review and Analysis. Personnel Psychology, Vol. 39, 24

1986, pp. 291-314. 25

21. Hayton, J. C., D. G. Allen, and V. Scarpello. Factor Retention Decisions in Exploratory 26

Factor Analysis: A Tutorial on Parallel Analysis. Organizational Research Methods, Vol. 27

7, No. 2, 2004, pp. 191-205. 28

22. de Winter, J. C., and D. Dodou. Factor Recorvery by Principal Axis Factoring and 29

Maximum Likelihood Factor Analysis as a Function of Factor Pattern and Sample Size. 30

Journal of Applied Statistics, Vol. 39, No. 4, 2012, pp. 695-710. 31

23. Anderson, J. C., and D. W. Gerbing. Structural Equation Modeling In Practice: A Review 32

and Recommended Two-Step Approach. Psychological Bulletin, Vol. 103, No. 3, 1988, 33

pp. 411-423. 34

24. Golob, T. F. Structural Equation Modeling for Travel Behavior Research. Transportation 35

Research Part B, Vol. 37, 2003, pp. 1-23. 36

25. Kaiser, H. F. The Application of Electronic Computers to Factor Analysis. Educational 37

and Psychological Measurement, Vol. XX, No. 1, 1960, pp. 141-151. 38

26. R Core Team, R: A Language and Environment for Statistical Computing, Vienna, 2014. 39


27. Rosseel, Y. Lavaan: An R Package for Structural Equation Modeling. Journal of Statistical 1

Software, Vol. 48, No. 2, 2012, pp. 1-36. 2

28. Hu, L.-t., and P. M. Bentler. Cutoff Criteria for Fit Indexes in Covariance Structure 3

Analysis: Conventional Criteria versus New Alternatives. Structural Equation Modeling: 4

A Multidisciplinary Journal, Vol. 6, No. 1, 1999, pp. 1-55. 5

29. Bandalos, D. L. Relative Performance of Categorical Diagonally Weighted Least Squares 6

and Robust Maximum Likelihood Estimation. Structural Equation Modeling: A 7

Multidisciplinary Journal, Vol. 21, No. 1, 2014, pp. 101-116. 8

30. Millsap, R. E. Statistical Approaches to Measurement Invariance. Routledge Taylor & 9

Francis Group, New York, 2011. 10

31. Millsap, R. E., and J. Yun-Tein. Assessing Factorial Invariance in Ordered-Categorical 11

Measures. Multivariate Behavioral Research, Vol. 39, No. 3, 2004, pp. 479-515. 12

32. Byrne, B. M., R. J. Shavelson, and B. Muthén. Testing for the Equivalence of Factor 13

Covariance and Mean Structures: The Issue of Partial Measurement Invariance. 14

Psychological Bulletin, Vol. 105, No. 3, 1989, pp. 456-466. 15

33. Wright, S. Path Coefficients and Path Regressions: Alternative or Complementary 16

Concepts? Biometrics, Vol. 16, No. 2, 1960, pp. 189-202. 17

34. Liao, F. H., S. Farber, and R. Ewing. Compact Development and Preference Heterogeneity 18

in Residential Location Choice Behaviour: A Latent Class Analysis. Urban Studies, Vol. 19

52, No. 2, pp. 314-337, 2015. 20

the importance of housing, accessibility, and ... paper.pdf · 39 constructs will significantly...

Documents